1. Field of the Invention
The invention relates to management of an image database and the retrieval of images therefrom.
2. Description of the Related Technology
The ability to search image databases and retrieve images therefrom with desired features or characteristics is important in many different environments. However, as a collection of images to be searched grows in size, the ability to search and evaluate the collection manually for images having the desired features becomes increasingly limited. It can be appreciated that huge image databases of thousands or even millions of images have been created which are essentially impossible to search and evaluate manually.
Several approaches have been used to automate the image search process. In some cases, images are digitized and stored in a database in association with one or more keywords which describe their content or character. Such a database can be searched linguistically for particular keywords, and images which are associated with these keywords are retrieved in response.
In a more recently developed alternative method, one or more “query images” are utilized, and images from the database which are in some sense similar to the query image are located and retrieved. In these systems, the pixel values of the query image and the images in the database are processed to produce a set of parameters indicative of color distribution, pixel intensity variation across the image, as well as other characteristics of the image as a whole. These parameters are calculated using various image filtering and processing techniques so as to produce a vector of feature parameters which is indicative of the image itself. The comparison process involves comparing feature vectors from images in the database with a query feature vector, and images from the database having similar feature vectors are retrieved. A system of this nature is described in U.S. Pat. No. 5,644,765 to Shimura et al., the disclosure of which is hereby incorporated by reference in its entirety.
The above described systems have several limitations. The most serious drawback for both cases is that image content is inadequately defined, which impacts both system recall and precision. Recall is the proportion of relevant images in the database that are retrieved, and precision is the proportion of retrieved documents that are actually relevant. These two measures may be traded off one for the other, and the goal of image retrieval is to maximize them both.